A few years ago, the only people I knew with "forward-deployed engineer" in their title worked at Palantir. This year it feels like every AI company is hiring them, and a growing number of the companies buying AI are too.
The numbers are hard to ignore. LinkedIn reported that forward-deployed engineering positions grew 42-fold between 2023 and 2025. LinkUp, which tracks job listings, counted about 300 open FDE roles through 2022 and 2023, and more than 5,000 by early 2026. An executive search firm told TechCrunch that the share of companies planning to hire FDEs went from 5–10% at the start of this year to 70% by the end of June.
Most of what I've read about this treats it as a career story. Good pay, new title, here's how to land the job. That's fine, but the more interesting question to me is why a role grows that fast in the first place. My answer: companies are hiring people to supply the flexibility their systems don't have.
What FDEs actually spend their time on
CIEL HR, which analyzed FDE hiring this summer, describes the job as integrating AI into existing technology environments and connecting models with business workflows. (If you want the basics on the role itself, Paolo Perrone has a solid explainer on HackerNoon already.)
That description is accurate. It's also a polite way of saying most of the work is plumbing and translation. Figuring out which of six systems actually owns the customer record. Writing glue code so a nightly batch export from 2011 can feed a model. Sitting with the ops team until you learn the three exception rules nobody ever wrote down. Very little of it touches the model.
IS research saw this coming
I study IT infrastructure flexibility, and this is the part that clicked for me. Back in 1995, Nancy Duncan described flexible infrastructure in terms of three technical properties: connectivity, compatibility, and modularity. Can the pieces reach each other? Can they share data that means the same thing on both sides? Can you swap one out without breaking the rest?
In 2000, Terry Anthony Byrd and Douglas Turner added a fourth: IT personnel flexibility. Their argument was that infrastructure includes people, and specifically how well those people understand both the technology and the business around it. A company with rigid systems and adaptable staff still has flexibility. It just keeps it in people's heads instead of in the architecture.
What I find telling is a follow-up Byrd and Turner published in 2001. They surveyed CIOs, mostly at Fortune 2000 companies, and found technical skills mattered most for infrastructure flexibility. Business knowledge, interpersonal skills and technology management didn't show the positive relationships they expected.
Now go read an FDE job posting. Yes, it wants a strong engineer. It also wants someone who will embed with the customer, learn their workflow, manage stakeholders and translate between engineers and the business. Those are the skills CIOs rated low 25 years ago, and the market now pays a premium for them. I don't think AI made those skills important. I think it showed how much enterprise systems had been quietly leaning on them all along.
The problem with human middleware
When systems can't connect, can't agree on what the data means, or can't change without breaking, somebody fills the gap by hand. An FDE is a very good somebody. In the short term that's often the fastest route to a working deployment, and I'm not knocking it.
But a person standing in for an interface behaves differently from an interface.
For one, it doesn't get reused. A clean API built for your first AI use case is still there for the fifth. A workaround that lives in one engineer's head only helps the next project if that engineer gets assigned to it.
It also leaves. If the FDE works for a vendor, what they learned about how your systems really fit together walks out when the contract ends. TechCrunch reported that some enterprises are now building internal FDE teams instead of relying on outside firms, partly to keep proprietary processes away from AI vendors. So at least some companies have noticed.
And it hides the cost. The pilot works, so the infrastructure gap never turns into a budget line. The next team pays for the same integration all over again.
Deloitte's 2026 State of AI in the Enterprise survey points the same way. Among more than 3,200 leaders, insufficient worker skills came out as the biggest barrier to integrating AI. Yet 84% of organizations hadn't redesigned jobs or workflows around it. Plenty of companies feel the people gap. Far fewer are changing how their people and systems fit together.
What I'd do with them instead
Hire FDEs if you need them. Early in a deployment, someone who's comfortable working in the mess is worth a lot. The real question is what you keep after they're done.
Start by asking them to write down every workaround: each manual mapping, one-off script, and undocumented rule they had to rely on. That list is the most honest map of your infrastructure gaps you'll ever get, and you've already paid for it.
Then watch for reuse. When your second and third AI projects kick off, check how much of the first project's integration work they could pick up. If the answer is close to none, you're paying for human middleware every single time.
If the FDEs are external, pair each with an internal counterpart, and ensure the code and documentation end up in your repositories rather than the vendor's.
Look inside, too. Your data engineers, analysts, and data scientists who already know the business domain are some of the cheapest FDEs you'll find. Give them time for integration work and credit for it, not just for shipping models.
Finally, take the workarounds that keep showing up and fund the boring fixes: real interfaces, shared data definitions, services you can change on their own. That's how flexibility moves out of people's heads and into the architecture.
"Isn't this just what new technology looks like?"
That's the fair pushback. AI is new, the interfaces aren't standard yet, and someone has to bridge the gap for now. For a first deployment, I agree. Even the people recruiting FDEs don't describe the role as permanent: the search firm founder TechCrunch spoke with said the job could disappear within 5 to 10 years. (His firm places FDEs, so read his numbers with that in mind.)
If the role is temporary, though, what it teaches you about your systems shouldn't be.
The signal worth paying attention to
I've started thinking of every FDE hire as a bug report about the infrastructure. It tells you exactly where your systems can't absorb something new on their own. You can keep renting that flexibility one engagement at a time. Or you can take what those engineers learn and build it in, so the next project needs fewer of them.



